CN109063537B - Hyperspectral image preprocessing method for unmixing of abnormal small target - Google Patents

Hyperspectral image preprocessing method for unmixing of abnormal small target Download PDF

Info

Publication number
CN109063537B
CN109063537B CN201810571359.2A CN201810571359A CN109063537B CN 109063537 B CN109063537 B CN 109063537B CN 201810571359 A CN201810571359 A CN 201810571359A CN 109063537 B CN109063537 B CN 109063537B
Authority
CN
China
Prior art keywords
pixels
pixel
spatial
hyperspectral image
unmixing
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810571359.2A
Other languages
Chinese (zh)
Other versions
CN109063537A (en
Inventor
邓宸伟
冯帆
唐林波
王文正
赵保军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Institute of Technology BIT
Shanghai Institute of Satellite Engineering
Original Assignee
Beijing Institute of Technology BIT
Shanghai Institute of Satellite Engineering
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Institute of Technology BIT, Shanghai Institute of Satellite Engineering filed Critical Beijing Institute of Technology BIT
Priority to CN201810571359.2A priority Critical patent/CN109063537B/en
Publication of CN109063537A publication Critical patent/CN109063537A/en
Application granted granted Critical
Publication of CN109063537B publication Critical patent/CN109063537B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB

Abstract

The invention discloses a hyperspectral image preprocessing method for unmixing of an abnormal small target. The method can solve the defect that the hyperspectral image preprocessing method serving for the unmixing task is easy to ignore the abnormal small target. The method fully aims at the characteristic of small target space scale, judges suspected targets on space dimension by using a sliding window, establishes similarity measurement weight according to neighborhood pixel positions, and discriminates the influence of different neighborhood pixels in a treatment window on the judgment of the space specificity of pixels to be detected; meanwhile, the suspected target is judged in the feature dimension by utilizing the feature that the target spectrum has specificity compared with the background spectrum and utilizing PCA (principal component analysis) conversion; and finally, the hyperspectral data is screened by combining a K-means method and an Orthogonal Subspace Projection (OSP) method, so that the data volume to be processed is effectively reduced, the unmixing precision is improved, and a great improvement space is provided in engineering application. The invention does not need to modify any subsequent end member extraction stage, and the algorithm is flexible to apply.

Description

Hyperspectral image preprocessing method for unmixing of abnormal small target
Technical Field
The invention relates to the technical field of remote sensing image processing, in particular to a spatial spectrum information combined hyperspectral image preprocessing method aiming at a task of unmixing a small spatial scale target (namely an abnormal small target) with abnormal spectral characteristics.
Background
The hyperspectral image has high spectral resolution and wide wave band range, and is widely applied to the military and civil fields. However, due to the limitation of its spatial resolution, spectra of different ground objects in the acquisition scene are mixed with each other, resulting in a mixed pixel. The mixed pixel decomposition is unique and important research content in the hyperspectral remote sensing technology, and has important significance in the aspects of hyperspectral image refinement processing and quantitative analysis. In order to improve the unmixing precision, unmixing algorithms based on different characteristics such as geometry, statistics and the like are continuously proposed, and the unmixing algorithms are respectively suitable for different scenes. However, many researches on hyperspectral unmixing only utilize the spectral information of the hyperspectral images, and spatial information contained in the hyperspectral images is not sufficiently explored, so that limitations are brought to unmixing precision. In order to solve the problem, a de-mixing algorithm fusing spatial spectrum information becomes a research hotspot, wherein a hyperspectral image preprocessing technology aiming at a de-mixing task can effectively screen candidate pixels without changing the existing de-mixing algorithm, so that the de-mixing precision is improved, and the operation burden is reduced, thereby being concerned.
In recent years, more and more hyperspectral image preprocessing techniques have been studied and proposed. However, when processing the hyperspectral image, the technologies usually do not consider the difference of the spectral characteristics and the spatial characteristics between different ground objects, uniformly perform undifferentiated processing, easily ignore abnormal small targets in the image, and bring difficulty to the task of image unmixing under the condition that the scene contains the abnormal small targets.
In the field of image preprocessing facing a hyperspectral image unmixing task, the method mainly comprises the following two types of methods: one is a spatial preprocessing method, which judges whether a pixel exists in a large-area similar region (i.e., a homogeneous region) of spectral characteristics in which an end member is more likely to appear by using spatial information contained in a hyperspectral image, sets a weight for the pixel according to the spectral similarity between the pixel and a neighborhood pixel, and guides the extraction process of the end member, thereby improving the unmixing precision. The method can enhance the end member search of the homogeneous region in the image, is suitable for the condition that the image has large homogeneous ground objects, but is not suitable for the condition that an abnormally small target exists. The other type is a spectrum preprocessing method, which mainly screens pixels in a space adjacent region according to spectrum characteristics for a subsequent end member extraction process. Under a linear mixed model, end members usually appear at the edge position of a spectral feature space, and the operation amount can be effectively reduced by primarily screening the end members according to the positions of pixels in the spectral feature space, but small targets with small space size and small occupied pixel number are easily missed in the screening process, so that the method is not suitable for the unmixing task under the condition of abnormal small targets.
Disclosure of Invention
In view of the above, the invention provides a hyperspectral image preprocessing method for unmixing of an abnormal small target, which can solve the problem that the hyperspectral image preprocessing method serving for the unmixing task is easy to ignore the abnormal small target.
The invention relates to a hyperspectral image preprocessing method aiming at unmixing of an abnormal small target, which comprises the following steps of:
step 1, extracting pixels of a suspected target:
step 1.1, extracting pixels of a suspected target from spatial dimensions:
performing sliding window processing on the hyperspectral image, respectively calculating the similarity degree between a central pixel and other neighborhood pixels in the sliding window, and setting the weight of the similarity degree between the neighborhood pixels and the central pixel according to the distance between the neighborhood pixels and the central pixel; taking the weighted sum of the similarity degrees of the central pixel and all the field pixels in the sliding window as the spatial specificity degree value of the central pixel; moving the sliding window to obtain the spatial specificity degree values of all pixels in the hyperspectral image; extracting the image elements of which the spatial specificity degree value is greater than or equal to a set threshold value A,form a suspected target image element set P1；
Step 1.2, extracting pixels of a suspected target from the characteristic dimension:
performing principal component analysis on the hyperspectral image to obtain a characteristic space matrix; respectively projecting all pixels in the hyperspectral image onto a feature space, extracting the largest 1-5% and the smallest 1-5% pixels in all projection values to form a suspected target image element set P2；
Step 1.3, taking P1And P2To obtain the pixel set P of the suspected targettarget；
Step 2, extracting background pixels:
step 2.1, clustering the hyperspectral images by adopting a K-means algorithm to obtain m regions and representative spectra of the regions;
2.2, selecting c most orthogonal region representative spectrums from the m region representative spectrums obtained in the step 2.1 by adopting an orthogonal subspace projection algorithm; the c regions which are most orthogonal to each other represent the pixels of the regions corresponding to the spectrums, namely a background pixel set is formed;
and 3, taking the pixel set of the suspected target obtained in the step 1 and the union set of the background pixel sets obtained in the step 2, namely the preprocessed hyperspectral pixel.
Further, in step 1.1, the spatial specificity degree of the central pixel is corrected, and the spatial specificity degree value ρ (i, j) of the central pixel r (i, j) is:
Figure BDA0001686044290000031
wherein α (i, j) is the weighted sum of the similarity degrees of the center pixel r (i, j) and all the domain pixels in the sliding window.
Further, the threshold a is:
A＝max_spatial-(max_spatial-min_spatial)×α
and the alpha is 5-10%, and the max _ spatial and the min _ spatial are respectively the maximum value and the minimum value in the spatial specificity degree values of all pixels in the hyperspectral image.
Further, in the step 1.2, a hyperspectral image pixel r is selectediProjection in a feature space
Figure BDA0001686044290000032
The pixels satisfying the following formula form a suspected target pixel set P2：
Figure BDA0001686044290000033
Or
Figure BDA0001686044290000034
Wherein β is 1% to 5%, and max _ project and min _ project are the maximum value and the minimum value of all projection values, respectively.
Further, in step 2.1, m is 2p, where p is the number of feature vectors set according to the scene complexity in the principal component analysis in step 1.2.
Has the advantages that:
the invention provides a hyperspectral image preprocessing method based on spatial-spectral information combination in the field of hyperspectral image preprocessing aiming at unmixing tasks, which aims at the unmixing tasks under the condition that abnormal small targets exist in scenes, solves the problems that the abnormal small target spectrums are easy to ignore and omit in the preprocessing process, and specifically comprises the following steps:
(1) the method of the invention fully aims at the characteristic of small target space scale, judges the suspected target in space dimension by using sliding window treatment, establishes similarity measurement weight according to the distance between the neighborhood pixels and the central pixel, and differentially treats the influence of different neighborhood pixels in the window on the judgment of the space specificity of the pixel to be detected, so that the method has robustness, effectively positions the suspected target from the space dimension and prevents omission.
(2) The method of the invention fully utilizes the characteristic that the target spectrum has specificity compared with the background spectrum, and utilizes PCA transformation to project the hyperspectral image to the characteristic space, thereby removing the redundancy of spectral information. And the suspected target is judged in the characteristic dimension, so that the noise interference can be effectively weakened, the calculated amount can be reduced, the suspected target can be positioned from the spectral dimension, and the omission can be avoided.
(3) The method provided by the invention fully utilizes the characteristic that the spectra of the lower end members of the unmixing task based on the linear mixing model have the maximum difference with each other, and combines the K-means method and the Orthogonal Subspace Projection (OSP) method to screen the hyperspectral data, thereby effectively reducing the data volume to be processed and improving the unmixing precision, and having a great promotion space in engineering application.
(4) The hyperspectral image spectral information and the spatial information are utilized and combined in the preprocessing stage, any modification is not needed in the subsequent end member extraction stage, and the algorithm is flexible to apply.
Drawings
FIG. 1 is a general flow chart of the practice of the present invention.
FIG. 2 is a schematic diagram of a sliding window for spatial specificity calculation according to the present invention.
Detailed Description
The invention is described in detail below by way of example with reference to the accompanying drawings.
The invention provides a hyperspectral image preprocessing method aiming at unmixing of an abnormal small target, and a flow chart of the method is shown in figure 1, and the method comprises the following steps:
the specific analysis is as follows:
step one, estimating the specificity of each pixel in the image on a spatial scale based on spatial information contained in a hyperspectral image, and judging a suspected target from the spatial dimension to obtain a suspected target pixel set P1。
The hyperspectral image processing method comprises the steps that sliding window processing can be carried out on a hyperspectral image by setting a certain window size, and similarity measurement of a central pixel and neighborhood pixels (other pixels except the central pixel in the sliding window) in the sliding window is set so as to measure the spectrum similarity degree between the central pixel and the neighborhood pixels; and adding weight for the similarity measurement according to the distance between the neighborhood pixels and the central pixel, calculating the weighted sum of the similarity measurement between each pixel and the neighborhood pixels thereof as the spatial specificity degree of the pixel, thereby differently treating the influence of different neighborhood pixels in a window on the spatial specificity judgment of the pixel to be detected, effectively positioning the suspected target from the spatial dimension, preventing omission and having robustness.
Specifically, the hyperspectral image to be processed is subjected to sliding window processing, and a specific spatial specificity calculation process is shown in fig. 2. Taking a pixel r (i, j) to be measured as a center, and determining the radius of a sliding window to be d ═ ws-1)/2 pixels, wherein ws is an odd number, defining the spatial specificity rho (i, j) of the center pixel r (i, j) in the sliding window as the weighted sum alpha (i, j) of the similarity measurement of the center pixel and the pixels in other fields in the sliding window, and establishing the similarity measurement weight according to the distance between the neighborhood pixel and the center pixel, namely the expression of the spatial specificity rho (i, j) of the center pixel r (i, j) is as follows:
Figure BDA0001686044290000061
wherein γ (z-i, s-j) ═ γ (r (z, s), r (i, j)), is a similarity measure between the central pixel r (i, j) and the neighborhood pixel r (z, s), and may be selected from euclidean distance or spectral angular distance; β (z-i, s-j) is the weight of the similarity measure value of the neighborhood pixel r (z, s) and the center pixel r (i, j), β (z-i, s-j) is proportional to the distance of the pixel r (z, s) and the center pixel r (i, j), and the weights of the similarity measure values of all neighborhood pixels are normalized so that the sum of the similarity measure value weights of all the neighborhood pixels is 1, i.e.:
Figure BDA0001686044290000062
further, to increase the difference in spatial specificity between different pixels, the calculation method of the spatial specificity degree value ρ (i, j) may be rewritten as:
Figure BDA0001686044290000063
and moving the sliding window to obtain a spatial specificity degree value of each pixel in the hyperspectral image, comparing the spatial specificity degree value of the pixel with a set threshold A, and if the spatial specificity degree value of the pixel is greater than or equal to the threshold A, judging the pixel to be a suspected target pixel. Recording the coordinates of the suspected pixel points with the space specificity degree larger than the threshold value A to obtain a suspected target pixel set P1。
The threshold a may be set according to the size of the target spatial difference, or may be set according to the maximum value max _ spatial and the minimum value min _ spatial of the obtained spatial specificity ρ (i, j) of all pixels:
A＝max_spatial-(max_spatial-min_spatial)×α
wherein alpha is 5-10%, the pixels screened out in this way are 5-10% of the pixels with the largest spatial specificity in the whole hyperspectral image, and the method accords with the characteristics of small space scale of an abnormally small target and obvious difference with surrounding neighborhood spectrums.
Secondly, performing Principal Component Analysis (PCA) transformation on the original hyperspectral image based on spectral information contained in the original hyperspectral image, projecting the original hyperspectral image to a feature space, analyzing the projection of each pixel of the original hyperspectral image in the feature space, performing suspected target judgment in a feature dimension, and obtaining a suspected target image metaset P2。
PCA is a basic hyperspectral data dimension reduction method, and the specific steps are as follows:
inputting a hyperspectral image
Figure BDA0001686044290000071
Wherein r isiThe number of the ith pixel in the hyperspectral image is 1,2,3, … and n, wherein n is the total number of the hyperspectral image pixels; and (3) calculating:
Figure BDA0001686044290000072
calculating the characteristic value of gamma and sequencing the characteristic values from large to small to obtain the characteristic value lambda1>λ2>…>λLAnd eachUnit characteristic vector corresponding to characteristic value
Figure BDA0001686044290000073
Then Q ═ a1,a2,…,aL]. And performing projection in a feature space on each pixel i-1, …, n to obtain:
Figure BDA0001686044290000074
most of the energy of the pixel is concentrated on the unit feature vector corresponding to the first largest feature values, so the unit feature vector corresponding to the first largest p feature values is selected according to the scene complexity. Calculating to obtain projection values of all pixels in the original hyperspectral image on a feature space consisting of p feature vectors
Figure BDA0001686044290000075
The maximum projection value max _ project and the minimum projection value min _ project on the feature space are recorded. According to the boundary position of the target pixel in the feature space of the data cloud consisting of all pixels, namely the target characteristic that the spectrum has specificity, the pixel P with the projection value of 1% -5% of the maximum projection value and 1% -5% of the minimum projection value on the feature vector is screened out2(ii) a That is to say if
Figure BDA0001686044290000076
Satisfies the following conditions:
Figure BDA0001686044290000077
or
Figure BDA0001686044290000078
Wherein, beta is 1% -5%, then the pixel r is considered asiThe projection positions on the feature space are positioned at two ends of the feature space and represent the pixel riData cloud consisting of all pixels in feature spaceI.e. its spectrum is specific, all r satisfying the above condition are selectediTo obtain a pixel set P2。
The pixels screened out in this way not only accord with the spectral characteristics of the boundary position of the end member in the characteristic space in the linear mixed model, but also meet the characteristic that the abnormal small target spectrum has specificity compared with most background spectra, and can effectively screen out suspected targets.
Step three, taking intersection of the pixel sets obtained in the step one and the step two:
Ptarget＝P1∩P2
obtaining a suspected target image element set P which has difference with surrounding neighborhoods and specificity in spectral dimensiontarget。
And step four, extracting a background pixel set in the suspected target pixel set.
Specifically, the original hyperspectral images can be clustered by using a K-means algorithm to obtain m regions, where the size of m is usually twice the number p of feature vectors in the two-step principal component analysis, that is, m is 2 p. Firstly, randomly selecting m pixels from n pixels as an initial clustering center; for the rest of other image elements, respectively allocating each image element to the most similar cluster (represented by the cluster center) according to the similarity (distance) between the image elements and the cluster centers; then calculating the clustering center of each new cluster (the mean value of all pixel spectra in the cluster); this process is repeated until the squared error is minimal. The square error is set as follows:
Figure BDA0001686044290000081
wherein theta isijIn pixel xi1 when classified into cluster j, and 0, mu otherwisejThe cluster center spectrum representing cluster j.
M regions are obtained by means of K-means, m being typically twice as large as the number p of eigenvectors in the two-principal component analysis step, i.e. m 2 p. Each region is composed of image elements with similar spectrums and can representIs composed of
Figure BDA0001686044290000082
Averaging all pixel spectrums in the region to obtain a representative spectrum M capable of representing the spectrum characteristics of the regionl，l＝1,2,...,m。
Figure BDA0001686044290000083
Wherein n islRepresenting the number of pixels in the ith area. Rl(i) The ith spectrum representing the l region.
And step five, utilizing an Orthogonal Subspace Projection (OSP) algorithm to perform projection selection on the region representative spectrums obtained in the step four, screening out c representative spectrums which are most orthogonal to each other, and reserving the c regions represented by the c representative spectrums as a background image element set.
Specifically, the c most orthogonal spectra are selected from all representative spectra according to the following projection rule, where the magnitude of c is typically set such that p ≦ c ≦ m. The area represented by the c most orthogonal representative spectra is the background area.
The selection process is as follows:
first, a first area B is selected1Is the region representing the most compact spectrum and can be determined by the following formula:
B1＝Rnum1
Figure BDA0001686044290000091
after obtaining the first region, the region representative spectrum U is recorded1＝[M1]Then, all the representative spectra of the m regions are projected using an orthogonal subspace projection operator, the region with the largest projection value being selected as the second region:
B2＝Rnum2
Figure BDA0001686044290000092
Figure BDA0001686044290000093
i represents the identity matrix, update U2＝[M1,M2]Repeating the above process, adding a new region each time:
Bj＝Rnumj
Figure BDA0001686044290000094
Figure BDA0001686044290000095
until the number of the selected regions meets the set number c, obtaining
Figure BDA0001686044290000096
Namely the background area image element set.
Step six, the suspected target image element set P obtained in the step three is usedtargetAnd the background area image element set obtained in the fifth step
Figure BDA0001686044290000097
Taking a union set:
Figure BDA0001686044290000101
and obtaining the preprocessed hyperspectral data.
Then, the preprocessed hyperspectral images can be subjected to unmixing to obtain targets (including small abnormal targets).
In summary, the above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (5)

1. A hyperspectral image preprocessing method for unmixing of an abnormal small target is characterized by comprising the following steps:
step 1, extracting pixels of a suspected target:
step 1.1, extracting pixels of a suspected target from spatial dimensions:
performing sliding window processing on the hyperspectral image, respectively calculating the similarity degree between a central pixel and other neighborhood pixels in the sliding window, and setting the weight of the similarity degree between the neighborhood pixels and the central pixel according to the distance between the neighborhood pixels and the central pixel; taking the weighted sum of the similarity degrees of the central pixel and all the field pixels in the sliding window as the spatial specificity degree value of the central pixel; moving the sliding window to obtain the spatial specificity degree values of all pixels in the hyperspectral image; extracting pixels with spatial specificity degree value greater than or equal to a set threshold value A to form a suspected target pixel set P1；
Step 1.2, extracting pixels of a suspected target from the characteristic dimension:
performing principal component analysis on the hyperspectral image to obtain a characteristic space matrix; respectively projecting all pixels in the hyperspectral image onto a feature space, extracting the largest 1-5% and the smallest 1-5% pixels in all projection values to form a suspected target image element set P2；
Step 1.3, taking P1And P2To obtain the pixel set P of the suspected targettarget；
Step 2, extracting background pixels:
step 2.1, clustering the hyperspectral images by adopting a K-means algorithm to obtain m regions and representative spectra of the regions;
2.2, selecting c most orthogonal region representative spectrums from the m region representative spectrums obtained in the step 2.1 by adopting an orthogonal subspace projection algorithm; the c regions which are most orthogonal to each other represent the pixels of the regions corresponding to the spectrums, namely a background pixel set is formed;
and 3, taking the pixel set of the suspected target obtained in the step 1 and the union set of the background pixel sets obtained in the step 2, namely the preprocessed hyperspectral pixel.
2. The hyperspectral image preprocessing method for unmixing of an abnormally small target according to claim 1, wherein in step 1.1, the spatial specificity degree of a central pixel is corrected, and the spatial specificity degree value ρ (i, j) of the central pixel r (i, j) is:
Figure FDA0001686044280000021
wherein α (i, j) is the weighted sum of the similarity degrees of the center pixel r (i, j) and all the domain pixels in the sliding window.
3. The hyperspectral image preprocessing method for unmixing of an abnormally small target according to claim 1, wherein the threshold a is:
A＝max_spatial-(max_spatial-min_spatial)×α
and the alpha is 5-10%, and the max _ spatial and the min _ spatial are respectively the maximum value and the minimum value in the spatial specificity degree values of all pixels in the hyperspectral image.
4. The hyperspectral image preprocessing method for unmixing of abnormally small targets according to claim 1, wherein in the step 1.2, a hyperspectral image pixel r is selectediProjection in a feature space
Figure FDA0001686044280000022
The pixels satisfying the following formula form a suspected target pixel set P2：
Figure FDA0001686044280000023
Or
Figure FDA0001686044280000024
Wherein β is 1% to 5%, and max _ project and min _ project are the maximum value and the minimum value of all projection values, respectively.
5. The hyperspectral image preprocessing method for unmixing of the abnormally small target according to claim 1, wherein in the step 2.1, m is 2p, where p is a feature vector number set according to scene complexity in the step 1.2 principal component analysis.
CN201810571359.2A 2018-06-06 2018-06-06 Hyperspectral image preprocessing method for unmixing of abnormal small target Active CN109063537B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810571359.2A CN109063537B (en) 2018-06-06 2018-06-06 Hyperspectral image preprocessing method for unmixing of abnormal small target

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810571359.2A CN109063537B (en) 2018-06-06 2018-06-06 Hyperspectral image preprocessing method for unmixing of abnormal small target

Publications (2)

Publication Number Publication Date
CN109063537A CN109063537A (en) 2018-12-21
CN109063537B true CN109063537B (en) 2021-08-17

Family

ID=64820414

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810571359.2A Active CN109063537B (en) 2018-06-06 2018-06-06 Hyperspectral image preprocessing method for unmixing of abnormal small target

Country Status (1)

Country Link
CN (1) CN109063537B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109858531B (en) * 2019-01-14 2022-04-26 西北工业大学 Hyperspectral remote sensing image fast clustering algorithm based on graph
CN112712028B (en) * 2020-12-30 2024-04-09 闽江学院 Spectrum unmixing method based on normalized ground object subspace projection

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103293523A (en) * 2013-06-17 2013-09-11 河海大学常州校区 Hyperspectral remote sensing small target detection method based on multiple aperture information processing
CN104331880A (en) * 2014-10-20 2015-02-04 西安电子科技大学 Hyper-spectral mixed pixel decomposition method based on geometric spatial spectral structure information
CN106600602A (en) * 2016-12-30 2017-04-26 哈尔滨工业大学 Clustered adaptive window based hyperspectral image abnormality detection method
CN106886760A (en) * 2017-01-24 2017-06-23 北京理工大学 A kind of EO-1 hyperion Ship Detection combined based on empty spectrum information
CN106919952A (en) * 2017-02-23 2017-07-04 西北工业大学 EO-1 hyperion Anomaly target detection method based on structure rarefaction representation and internal cluster filter
CN107274416A (en) * 2017-06-13 2017-10-20 西北工业大学 High spectrum image conspicuousness object detection method based on spectrum gradient and hierarchical structure
CN108073895A (en) * 2017-11-22 2018-05-25 杭州电子科技大学 A kind of EO-1 hyperion object detection method based on the mixed pretreatment of solution

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8289513B2 (en) * 2009-05-01 2012-10-16 Chemimage Corporation System and method for component discrimination enhancement based on multispectral addition imaging
JP2013093685A (en) * 2011-10-25 2013-05-16 Sumitomo Electric Ind Ltd Imaging apparatus

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103293523A (en) * 2013-06-17 2013-09-11 河海大学常州校区 Hyperspectral remote sensing small target detection method based on multiple aperture information processing
CN104331880A (en) * 2014-10-20 2015-02-04 西安电子科技大学 Hyper-spectral mixed pixel decomposition method based on geometric spatial spectral structure information
CN106600602A (en) * 2016-12-30 2017-04-26 哈尔滨工业大学 Clustered adaptive window based hyperspectral image abnormality detection method
CN106886760A (en) * 2017-01-24 2017-06-23 北京理工大学 A kind of EO-1 hyperion Ship Detection combined based on empty spectrum information
CN106919952A (en) * 2017-02-23 2017-07-04 西北工业大学 EO-1 hyperion Anomaly target detection method based on structure rarefaction representation and internal cluster filter
CN107274416A (en) * 2017-06-13 2017-10-20 西北工业大学 High spectrum image conspicuousness object detection method based on spectrum gradient and hierarchical structure
CN108073895A (en) * 2017-11-22 2018-05-25 杭州电子科技大学 A kind of EO-1 hyperion object detection method based on the mixed pretreatment of solution

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
A Discriminative Metric Learning Based Anomaly Detection Method;Bo Du;《IEEE Transactions on Geoscience and Remote Sensing》;20141231;全文 *
LOW-RANK AND SPARSE TENSOR RECOVERY FOR HYPERSPECTRAL ANOMALY DETECTION;Dai Jiahui;《2017 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM》;20171231;全文 *
面向异常检测的多源遥感影像融合技术研究;钟圣唯;《中国优秀硕士学位论文全文数据库 信息科技辑》;20160215;全文 *
高光谱影像集成学习分类及后处理技术研究;徐卫霄;《中国优秀硕士学位论文全文数据库 信息科技辑》;20120715;全文 *

Also Published As

Publication number Publication date
CN109063537A (en) 2018-12-21

Similar Documents

Publication Publication Date Title
KR101117837B1 (en) Multi-image feature matching using multi-scale oriented patches
US6990233B2 (en) Apparatus and method for extracting object based on feature matching between segmented regions in images
KR101247147B1 (en) Face searching and detection in a digital image acquisition device
US7734064B2 (en) Method and apparatus for classifying geological materials using image processing techniques
CN107194408B (en) Target tracking method of mixed block sparse cooperation model
CN114418957A (en) Global and local binary pattern image crack segmentation method based on robot vision
CN111444948B (en) Image feature extraction and matching method
CN109063537B (en) Hyperspectral image preprocessing method for unmixing of abnormal small target
CN104978738A (en) Method of detection of points of interest in digital image
CN113269201A (en) Hyperspectral image band selection method and system based on potential feature fusion
CN110991493A (en) Hyperspectral anomaly detection method based on collaborative representation and anomaly elimination
CN110852207A (en) Blue roof building extraction method based on object-oriented image classification technology
CN112164093A (en) Automatic person tracking method based on edge features and related filtering
US8126275B2 (en) Interest point detection
US20230144724A1 (en) Method for Finding Image Regions that Significantly Influence Classification in a Tool for Pathology Classification in a Medical Image
CN107346549B (en) Multi-class change dynamic threshold detection method utilizing multiple features of remote sensing image
CN110751671B (en) Target tracking method based on kernel correlation filtering and motion estimation
US20110135181A1 (en) polynomial fitting based segmentation algorithm for pulmonary nodule in chest radiograph
CN112991425B (en) Water area water level extraction method and system and storage medium
CN115205251A (en) Method for evaluating geometric quality availability of optical remote sensing image
CN115511928A (en) Matching method of multispectral image
CN110263777B (en) Target detection method and system based on space-spectrum combination local preserving projection algorithm
CN117152163B (en) Bridge construction quality visual detection method
US20110091106A1 (en) Image Processing Method And System
US20220414827A1 (en) Training apparatus, training method, and medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
TA01 Transfer of patent application right
TA01 Transfer of patent application right

Effective date of registration: 20200116

Address after: 100081 No. 5, Zhongguancun South Street, Haidian District, Beijing

Applicant after: BEIJING INSTITUTE OF TECHNOLOGY

Applicant after: Shanghai Satellite Engineering Research Institute

Address before: 100081 No. 5, Zhongguancun South Street, Haidian District, Beijing

Applicant before: BEIJING INSTITUTE OF TECHNOLOGY

GR01 Patent grant
GR01 Patent grant